Update transformer.py

2024-08-01 11:17:07 +03:00
parent 9c7322184f
commit b99417fd4a
1 changed files with 1 additions and 1 deletions
@@ -331,7 +331,7 @@ class RoPEAttention(Attention):
        # Attention
        if not OLD_TORCH:
            if not MATH_KERNEL_ON and OLD_GPU and dropout_p > 0.0:
-                backends.append(SDPBackend.MATH))
+                backends.append(SDPBackend.MATH)
            with sdpa_kernel(backends):
                out = F.scaled_dot_product_attention(q, k, v, dropout_p=dropout_p)
        else: